無建背景模式即時物件向量偵測於DSP實作及人行計數之應用

Real-Time Non-Background Motion-Based Object Detection in DSP Implementation and Application for People Counting

高義舜(Yi-Shun Kao)

指導教授 ： 林道通

國立臺北大學/電機資訊學院/電機工程研究所/碩士(2010年)

摘要

隨著安全與隱私的日漸重視，視訊監控的需求也日漸增加。而在監控系統中，最基礎也最重要的是物件偵測技術，尤其是在很複雜繁忙的環境下。本論文提出一種無建背景式即時物件向量偵測演算法，並將所提出的演算法移植至低成本的DSP嵌入式平台，以及將其延伸應用，發展成人行計數系統。此向量偵測方法主要在於利用三步搜尋法來指定方向的向量，並且再透過新的向量分析演算法來分析。若是有物件順著我們指定的方向穿越過我們事先定義的閘門，我們的系統將會偵測出此物件。此外我們也改進與最佳化此向量偵測演算法，然後移植至德州儀器公司的Davinci DM6446和DM DSP EVM。在最後我們也將此方法擴展應用，此方法配合其他像是3 successive-frames differencing等演算法來建構出人行計數系統。在實驗結果中可看出我們所提出的方法能即時運行，而且也具有相當高的偵測準確率。

關鍵字

視訊監控,向量評估,三步收尋法,移動向量分析,最佳化方法,三張影像差分法

並列摘要

To ensure public and private safety and security, the demand of intelligent video surveillance has become overwhelmingly increasing. One of the fundamental and important processes of the surveillance system is motion detection especially under very complex situations. The objective of this thesis is to present a non-background motion based object detection algorithm, and to implement the proposed algorithm on a low-cost DSP platform and to extend it to the application of people counting. The proposed motion detection method convolves the estimation of the appointed motion with three step search and motion analysis by a novel motion vector analysis algorithm to detect moving objects passing through a pre-defined gate along the desired direction. We further improve and optimize this motion detection algorithm, and then implement it on TI Davinci DM6446 and DM6437 DSP EVM boards. Besides, in the last part for this thesis shows the extended application of people counting on PC by this algorithm along with 3 successive-frames differencing method and others. The experimental results reveal that the final system works in real-time and performs excellent correct detection accuracy.

並列關鍵字

Video surveillance, motion estimation, Three step search, motion vector analysis, optimization, DM6437, DM6446, 3 successive-frames differencing

參考文獻

[1] Texas Instruments. sprue66c-DVEVM Getting Started Guide, March 2007.

[4] Cheng-Hui Wu. An Efficient Moving Shadow Removal Algorithm and Its Application of Traffic Flow Detection. 2009.

Tracking of the Human Body. 19:780-785, 1997.

[7] A.L. Hironobu, A.J. Lipton, H. Fujiyoshi, and R.S. Patil. Moving Target Classification and Tracking From Real-Time Video. 1998.

[11] C.H. Cheung and L.M. Po. A novel cross-diamond search algorithm for fast block motion estimation. IEEE Transactions on Circuits and Systems for Video Technology, 12(12):1168-1177, 2002.